SP500215 TREC-II Routing Experiments with the TRW/Paracel Fast Data Finder chapter M. Mettler National Institute of Standards and Technology D. K. Harman 6.0 Conclusions and Future Plans Overall, our results for `fl[OCRerr]EC-2 are encouraging. We were successfull in adapting the FDF hardware to running a soft statistical information retrieval algorithm. Not only did the FDF run all the final searches with no significant post-processing, we also found it a useful tool for experimenting with different proximity windows and deriving database frequencies for phrases and special features. We plan to continue our work and to participate in TREC-hI next year. The lessons we learned from TREC-I have proven very valuable, and so too should the lessons of TREC-Il. For TREC-hI, we will continue to examine the range of available query formulation, execution and evaluation models, with an eye to adapting those models for use with the FDF search hardware. We are looking at methods to use the raw horsepower of the FDF to expand existing techniques in ways which had previously been considered too computationally expensive. We are also struck by the observation that different query generation techniques are effective on different topics. A system that could execute a range of methods might be able to match the query generation approach to the topic type. We plan to consider possible schemes for characterizing the topic in advance so that the system might be able to use the best method. 208